Hal 9000 and AI

Home , HAL 9000, Monolith (Space Odyssey)

Chapter 12 Hal 9000 and AI

Hal 9000, an intelligent computer system, plays an important part in the novel “2001: A Space Odyssey” written by Arthor Clark, published for the ﬁrst time in 1968.

This novel was later made into a ﬁlm, directed by Stanley Kubrick, also in 1968.

The following ﬁgure shows the eyes of Hal 9000.

1 What is the story?

Below is a much simpliﬁed plot summary of this ﬁlm, emphasizing the role played by Hal 9000:

“When the world is ruled by apes, one particular group discovers a mysterious rectangular monolith near their home.

Astronaut David Bowman, along with four companions, sets oﬀ for Jupiter on a spaceship controlled by HAL 9000, a revolutionary computer system that is every bit mankind’s equal, and perhaps his superior.

When HAL endangers the crew’s lives for the sake of the mission, Bowman will have to ﬁrst overcome the computer, then travel to the birthplace of the monolith.”

2 What does Hal 9000 do?

The following is a scenario as depicted in the novel, where Hal 9000 tries to take control of the spaceship:

“Hal, switch to manual hibernation control.”

Question: How does Hal 9000 understand Dave’s command? What kind of ability should Hal possess to distinguish diﬀerent people’s voice?

“I can tell from your voice harmonics, Dave, that you’re badly upset. Why don’t you take a stress pill and get some rest?”

Question: How can Hall 9000 tell Dave is upset from the harmonics of his voice? Moreover, now can Hal make the prescription for Dave? 3 “Hal, I am in command of this ship. I order you to release the manual hibernation control.”

Question: How does Hal tell the diﬀerence between “order” and “request”?

“I’m sorry, Dave, but in accordance with special subroutine C1435-dash-4, quote, When the crew are dead or inca- pacitated, the onboard computer must assume control, unquote. I must, therefore, overrule your authority, since you are not in any condition to exercise it intelligently.”

Question: What is “C1435-dash-4”? How does Hal remember its content? How does Hal ﬁnd such a special routine that is immediately applicable? Is the reasoning that Hal just made sound? Is it rule based? 4 “Hal,” said Bowman, now speaking with an icy calm. “I am not incapac- itated. Unless you obey my instruc- tions, I shall be forced to disconnect you.”

Question: Can Hal tell the “icy” tone of Bow- man?

“I know you have had that on your mind for some time now, Dave, but that would be a terrible mistake. I am so much more capable than you are of supervising the ship, and I have such enthusiasm for the mission and conﬁ- dence in its success.”

Question: How can Hal know that Bowman has this tough in his mind for a while? How does Hal rich the conclusion that Bowman might make a terrible mistake? Will you do the same? 5 “Listen to me very carefully, Hal. Un- less you release the hibernation control immediately and follow every order I give from now on, I’ll go to Central and carry out a complete disconnection.”

Hal’s surrender was as total as it was unexpected. “O.K., Dave,” he said. “You’re certainly the boss. I was only trying to do what I thought best. Nat- urally, I will follow all your orders. You now have full manual hibernation control.”

Question: What is the Achilles heel of Hal 9000? Is this really fatal to Hal 9000? What happens if he carries battery? How do we control such a powerful companion if the need arises?

6 Hal 9000’s ability

It is obvious that Hal 9000 is able to “listen”, “think”, and “talk” to people in real time, an “intelligent agent”, just like us. On the other hand, Hal is man made. Thus, we call such an entity a one with artiﬁcial intelligence.

All these capabilities were mostly pure science ﬁction. Indeed, back in 1966, two years prior to the publication of this novel, researchers Bhi- mani, B., Merrill, R., Mitchell, R., and Stark, M., suggested that a person sitting at a desk, by means of a small digitizer and a telephone, could communicate with a database on a mainframe.

7 What is AI?

Roughtly speaking, Artiﬁcial Intelligence is in- tended to understand, and build, intelligent entities.

AI has produced many interesting and important results. But, it used to be diﬃcult.

For example, in the 1960’s, on then a very powerful IBM 360 mainframe, an analysis of a sample required 85 seconds for each second of the voice sample being translated, too slow to help Hal to communicate with his human colleagues in real time.

8 We have achieved quite a bit....

We have made quite a progress along this line: speech recognition and synthesis, e.g., the Siri system equipped with IPhone 4S; logic reasoning and/or planning, e.g., IBM Deep Blue; machine learning, e.g., IBM Watson system; intelligent agents, e.g., automatic vehicle.

For example, in the Third DARPA Grand Chal- lenge took place on November 3, 2007, an au- tonomous vehicle built by a CMU Racing Team completed the 55 mile course in an Urban set- ting, Victorville, California, in about four hours and won a USD 2M prize.

Let’s have a look....

9 Why are we interested in AI?

With AI, we want to study how it is possible for a brain to perceive, understand, predict, and then manipulate the world, which is much larger and complicated than itself.

A much more diﬃcult question is that, if we understand such a process, how could we build something with these features?

With its fast speed and big memory, Comput- ers provide an ideal platform for us to test out our assumptions and realize our dreams.

10 Both the newest and the oldest

AI is one of the newest discipline. It was for- mally initiated in the mid 1950’s. Although it has accomplished quite a bit, there is still a long way to go.

On the other hand, the study of intelligence is also one of the oldest. For 2,000 years, philosophers have tried to understand how see- ing, learning, remembering, and reasoning, could, as well as should, be done.

The emergence of computers in the early 1950’s made speculations into a real experimental and theoretical science.

11 Computer and AI

With its huge memory and stunning speed, be- sides providing a vehicle for creating artiﬁcial intelligence entities, the computer also provides a tool for testing theories of intelligence, many of which could not stand the test, just like many of the physics theories.

AI currently consists of many sub-ﬁelds. From such general areas as logic reasoning, to some very speciﬁc ones, such as playing chess.

Many scientists of other areas often move into AI, where they ﬁnd the theories and tools that systemize what they have been doing for a long time; while AI workers may choose to apply their methods to any area of human intelligence.

12 Various perspectives

There are various deﬁnitions of artiﬁcial intelligence. Some of them emphasize more on the thought process and reasoning; while others emphasize on the behavior.

Another angle is that if it tries to character- ize human performance; or some ideal performance, i.e., if it always does the right thing, or, rationally.

Thus, there could be four approaches, which have all been followed. A human-centered approach must be an experimental one, involving assumptions, and experimental conﬁrmations; while a rationalist approach involves a combi- nation of mathematics and engineering.

13 Turing (1912-1954) proposed a test to provide an operational definition of intelligence, when he defined intelligent behavior as the ability to achieve human-level performance in all cognitive tasks, sufficient to fool an interrogator. More specifically, a human interrogator asks questions to, and get answers from, two sources, one is controlled by a computer, the other by a human being, without knowing which is which.

If the interrogator cannot distinguish the sources based on the answers, then we may say the compute exhibits suﬃciently human intelligence to pass the Turing test. What does it take to pass?

The Loebner Prize, that started in 1990, is an annual competition that awards prizes to the Chatterbot considered by the judges to be the most human-like among the competitors. The format of the competition is that of a standard Turing test.

The 2010 Loebner Prize Competition in Arti- ﬁcial Intelligence was held on October 23 at California State University, Los Angeles. with the top prize of $3,000 and a bronze medal.

To pass such a test, a computer would have to possess such abilities as natural language pro- cessing, knowledge representation, automated reasoning, and machine learning.

Who won the Loebner Test this year?

14 Think humanly: Cognitive modeling

Following this approach, we have to set up a theory on human being’s thinking ﬁrst by studying our own thinking process, or via psy- chological experiments.

Then, we have to express the theory as a computer program. If the program’s I/O and tim- ing behavior match with the real thing, we will have some evidence that some of the program’s mechanisms may also be operating in humans.

15 The ﬁled of cognitive science brings together computer models of AI and experimental tech- niques of Psychology to construct precise and testable models of the workings of human mind.

The GPS(General Problem Solver) program (1957, Simon and Newell) is to compare its reasoning steps with that of human subjects solving the same problems.

16 Think rationally: The laws of thought

Aristotle(384-322 BC) was one of the ﬁrst to attempt to formalize the thinking process. His famous syllogisms provides pattern for argument structures that always gave correct con- clusions given correct premises. E.g., “Socrates is a man; all men are mortal; therefore Socrates is mortal.”

By 1965, programs came to life which, taking a problem represented in logic notation, is guar- anteed to ﬁnd a solution, if there is one, given enough space and time. The trick is that, if no solution exists, we might never know it.

17 Theory and practice

It is not easy to covert informal knowledge into formal notation(How does a baby recognize her mother’s face in just a few days?) Also, there is a big gap between being able to do it in principle and solve it in practice.

In fact, two essential issues in AI is knowledge representation and search. The former deals with representing what we know so that computers also “know”, and the second is how to let a computer ﬁnd the useful information to solve a problem at hand.

18 Act rationally: The rational agent

It is to act to achieve one’s goals, given one’s beliefs. An agent is just something that per- ceives and acts. In this approach, AI is regarded as the study and construction of rational agents.

Although making correct inferences is some- times part of being a rational agent, it is not always the case, since there might not be any provably correct things to do; or it is simply an act of gut feeling, e.g., pulling one’s hand oﬀ a hot stove.

All the cognitive skills listed under Turing Test are needed to allow rational actions.

19 This approach has two advantages. It is more general than the “laws of thought” approach since correct inference is only a useful mecha- nism, but not necessary.

It is also more amenable to scientiﬁc devel- opment than human centered approaches, be- cause the concept of rationality is clearly de- ﬁned and completely general.

The Grand Challenge might be regarded as a good test of this theory.

20 Philosophical foundation

At around 450 B.C., Aristotle, trying to for- mulate the laws that governs the rational part of the mind, developed an informal system of syllogism for proper reasoning, which in principle allowed one to mechanically generate conclusion, given initial premises. He also noted intuitive reason.

With at least part of the mind understood, the next step is to consider the mind as a physical system. Descartes (1596–1650) told the difference between mind and matter, as well as the associated problems.

One problem with a purely physical conception of the mind is that it leaves little room for free will. Although Descartes strongly believed the power of reasoning, he also believed that part of the mind is exempt from the physical laws. Thus, the dualism. 21 On the other hand, materialist held that all the world operates according to certain physical laws. In between there could be an in- termediate position, in which one accepts that the mind has a material basis, but denies that it can be explained by a reduction to ordinary physical processes.

By and large, philosophy laid a tradition in which the mind was conceived of as physical device operating mainly by reasoning with the knowledge it has.

22 The next problem is that where does knowledge come from?

Eventually, Rusell (1872–1970) suggested that all knowledge can be characterized by logical theories connected to observation sentences corresponding to sensory inputs.

Conﬁrmation theorists also tried to understand the process of acquiring knowledge from expe- rience.

This is perhaps where Expert System kicks off. Through knowledge engineering, we get the knowledge out of an expert, such as how to fix a car, how to put compatible parts together to assemble a computer, how to find why some one is sick, ..., and represent those knowledge to the computers, so that computer becomes an expert. 23 The last element of philosopher’s perception of the mind is the connection between knowledge and action. Particularly, how can an action be justified?

A typical technique in this regard is the means- ends analysis as put forward in GPS program by Newell and Simon around 1970’s. The following argument is fairly common:

I want to get an ’A’ in this course. The difference between what I have (nothing) and what I want is one of a number, at least 92.5. To reach that goal, I have to do well in the following areas: homework, projects, midterm, and final. Thus the ends of getting of an A justifies my action of working hard in reading the text books, staying in programming lab until 2 a.m., bugging the professor in his office the whole time, etc..

24 Activities

Choose a topic and write a one-page, single space, essay.

• What does the character of Hal 9000 represent? Discuss the possible issues and their implications when human beings are able to come up with such an intelligent prod- uct(s).

• Explore the status quo of artiﬁcial intelligence at present. What are among the most successful accomplishment of mod- ern AI technology?

• What are some of the other inﬂuential AI related characters as described in science ﬁction novels?

25 • Play with Eliza, do some research about its history and technology behind it. Why can Alice talk with you?

• What is the main diﬀerence between a robot and human being?

• If you have watched Spielberg’s “Artiﬁcial Intelligence:AI”, why does David want to be a real boy? What did he go through? Compare this David ﬁgure with Hal 9000.

• Have you used, or seen a demo, of iRobot, the carpet cleaner? What kind of intelligence do you think it shows? What kind of issues will be involved in making a prod- uct that will automatically cut your lawn?

26 • What is involved in an automatic vehicle? What issues have to addressed before putting it on road? Any commercial attempts have been made? What is the status?

• What do you think about the Turing Test? Do you think a program really demonstrates intelligent behavior if it passes the Turing test? Why or why not?

• Do some investigation of AI application in your discipline?

• What do you think will happen to AI in the future?